A sample statistic is a number that describes a measurable characteristic of a sample, which is a subset of a larger population. If you survey 1,000 voters and find that 54% prefer a particular candidate, that 54% is a sample statistic. It summarizes your data, and it serves as your best estimate of what the full population of all voters actually thinks. The concept is foundational to almost everything in statistics, from political polls to medical trials.
Statistics vs. Parameters
The distinction between a statistic and a parameter is one of the first things you encounter in any statistics course, and it comes down to scope. A parameter describes an entire population. A statistic describes a sample drawn from that population. If you could somehow measure every single member of a population, you’d have a parameter. Since that’s rarely possible, you measure a sample instead and get a statistic.
Here’s a simple test: Does the number come from a complete count of every member of a group? If yes, it’s a parameter. If it comes from a portion of that group, it’s a statistic. The average height of every student enrolled at a university, calculated from the full roster, is a parameter. The average height of 200 randomly selected students from that university is a statistic.
The whole point of calculating a sample statistic is to estimate the unknown population parameter. This process is called point estimation. Your sample average is a point estimate of the true population average. Your sample proportion is a point estimate of the true population proportion. The statistic is what you can measure; the parameter is what you actually want to know.
Common Sample Statistics and Their Notation
Statistics uses a consistent naming convention: Latin (ordinary) letters for sample statistics, Greek letters for population parameters. Once you know this rule, notation becomes much easier to read.
- Mean: The sample mean is written as x̄ (“x-bar”). The population mean is μ (“mu”). Both represent the average value.
- Standard deviation: The sample version is s. The population version is σ (“sigma”). Both measure how spread out values are from the mean.
- Variance: Sample variance is s². Population variance is σ². Variance is just the standard deviation squared.
- Proportion: The sample proportion is p̂ (“p-hat”) or sometimes just p. The population proportion is π (“pi”) or P.
- Correlation coefficient: The sample version is r. The population version is ρ (“rho”). Both measure the strength of a linear relationship between two variables.
You don’t need to memorize all of these to understand the core idea. The key takeaway is that every population parameter has a corresponding sample statistic, and statisticians use different symbols to keep them clearly separated.
Why Sample Statistics Vary
If you draw two different random samples from the same population, you’ll almost certainly get two different values for your statistic. Survey 1,000 voters today and get 52% support for a candidate; survey a different 1,000 voters tomorrow and you might get 49%. Neither sample is “wrong.” This natural fluctuation is called sampling variability (sometimes called sampling error, though “error” is a bit misleading since nobody made a mistake).
Sampling variability is the gap between your sample statistic and the true population parameter, and it exists purely because you’re looking at a subset rather than the whole. It’s the reason political polls conducted during the same week can produce noticeably different results. In October 2012, a Gallup poll showed Mitt Romney leading Barack Obama by 7 points among likely voters, while a University of Connecticut poll covering nearly the same time period showed Obama ahead by 3 points. Both polls used samples of roughly 1,000 to 2,000 people, and sampling variability explains much of the discrepancy.
This is also why polls report a margin of error. That margin is a statistical measure of how much the sample result could differ from the result you’d get if you interviewed every likely voter in the country.
How Sample Size Affects Accuracy
Larger samples produce more reliable statistics. This is intuitive: asking 10 people about their voting preference gives you a rougher picture than asking 10,000. But the relationship isn’t linear. Going from 100 to 1,000 respondents makes a dramatic difference in precision. Going from 10,000 to 11,000 barely moves the needle.
As sample size increases, the confidence interval around your statistic gets narrower. A confidence interval is the range within which the true population parameter likely falls. With a small sample, that range is wide because there’s more uncertainty. With a large sample, it tightens considerably. The actual estimate (the statistic itself) tends to stay relatively stable once you have a reasonable sample size, but your certainty about that estimate improves.
Statistical power also increases with sample size. Power is the ability to detect a real effect when one exists. A study with too few participants might miss a genuine difference between two groups simply because the sample wasn’t large enough to distinguish a real signal from random noise.
What Makes a “Good” Statistic
Not all sample statistics are equally useful as estimates of population parameters. Statisticians evaluate estimators based on a key property: bias. An unbiased estimator is one whose average value, across many hypothetical samples, equals the true population parameter. The sample mean is an unbiased estimator of the population mean. If you could draw thousands of samples and calculate x̄ each time, the average of all those x̄ values would converge on μ.
Bias matters because a systematically off-target statistic will mislead you no matter how large your sample gets. An unbiased statistic might be a little high in one sample and a little low in another, but it won’t consistently land on the same wrong side. This is one reason the sample mean is so widely used: it naturally centers on the population mean without any correction needed.
Beyond bias, precision also matters. Two statistics can both be unbiased, but one might bounce around wildly from sample to sample while the other stays relatively tight. The tighter one is more useful in practice because any single estimate you calculate is more likely to be close to the truth.
Everyday Examples
Sample statistics show up constantly in daily life, even when they aren’t labeled as such. When a news report says the average American household carries a certain amount of credit card debt, that number almost certainly comes from a sample survey, not a census of every household. It’s a sample statistic being used as an estimate of the population parameter.
Medical research works the same way. A clinical trial might enroll 5,000 participants and find that a new drug reduces blood pressure by an average of 8 points. That 8-point reduction is a sample statistic. Researchers use it to estimate how much the drug would reduce blood pressure across the entire population of people with similar conditions. The confidence interval around that estimate tells you how much wiggle room exists.
Quality control in manufacturing relies on sample statistics too. A factory producing thousands of bolts per day doesn’t measure every single one. Instead, workers pull a sample, measure the average diameter and its variability, and use those statistics to decide whether the production line is running within acceptable tolerances.